Explore the Power of External Data in Denoising Task

The goal of this paper is to explore the power of external data in the image denoising task, that is, to show that with taking advantage of an immense amount of information provided by external datasets, external denoising method should be more promising than internal denoising method which only extracts information from the input noisy image itself. In this paper, we present a simple external denoising method which combines Non Local Means (NLM) [Buades et al. 2005] with a randomized patch matching algorithm [Barnes et al. 2009] to denoise the input image (with an large enough external dataset) efficiently. Experimental results on a large set of images demonstrate that this external denoising method can outperform the according internal NLM and be competitive with the method [Mosseri et al. 2013] which properly combine the denoising results of both internal and external NLM. However, one drawback of the external denoising method is that compared with internal method, it is more vulnerable to noise overfitting problem. At the end of the paper, we also discuss a possible extension — applying adaptive patch size during denoising to reduce the overfitting problem and to make the external denoising method even more powerful.